What you connect. What Jigward can reach.

Organization-scoped user requests check membership and roles before using the organization’s data. Background work is also bound to an organization. Within it, a Task or Duty works in its own Context.

Every member can access every Context in their organization. Contexts group work; they do not separate those members into private teams. A private Chat is readable only by its member, although product knowledge intended for the Context can be retained separately.

• How Contexts divide work
• How private Chats work

To answer or act, Jigward can send relevant repository content, project knowledge, messages, tool results, logs and page content for inference. This information can be confidential. Keeping a credential out of the sign-in tool’s output does not anonymize the project.

Models are served through Amazon Bedrock in US Regions. AWS states that model providers do not have access to Bedrock customer prompts and completions.

Retention depends on the model and account settings. Fable’s standard AWS policy permits retention of inputs and outputs for up to 30 days for abuse detection, with possible human review of flagged traffic. AWS documents exceptions for eligible customers. Jigward does not promise universal zero retention.

• AWS: data protection in Bedrock
• AWS: model-specific abuse detection and retention

Custom API

Jigward’s backend authenticates the configured operation. Stored custom API credentials are encrypted with AES-256-GCM using data keys protected by AWS KMS, bound to the organization, purpose and integration. Known credential values are masked from the API’s text response.

Browser sign-in

The model refers to a saved field by its name, such as PASSWORD. Jigward fills the value itself, only on the profile’s configured sites; the fill tool does not return the value to the model. Stored browser profile values are not yet covered by the application-level KMS encryption used for custom API credentials.

CloudWatch Logs

An admin grants a role in your AWS account access to the log groups you allow. Jigward assumes that role with temporary credentials when needed, without storing a permanent AWS access key from you.

These mechanisms are not universal secret detection. Secrets already embedded in code, logs or page content can still reach a model. A screenshot can capture a credential that a page displays.

Work that needs execution gets a disposable AgentCore microVM workspace, removed when the work ends and not reused for later work. Jigward does not inject its integration credentials into it. Repository changes are published by Jigward’s backend.

A workspace can currently reach the public internet. Restricted outbound networking and a dependency mirror are planned, not active protections. A disposable workspace is not a promise of complete network isolation.

Jigward stores project knowledge, conversations and records of its work. A Task’s history can include commands, test output, changes, log extracts, screenshots and browser recordings. Reports go back to the channel where the work was given, so that channel’s audience can see the evidence shared there.

Bedrock retention and Jigward’s own storage are separate. Platform-wide retention periods, backup deletion and data-export arrangements are not published commitments today. Discuss your data requirements before connecting a project in closed testing.

Repository changes come back as draft pull requests for a person to merge. You can stop a Task; stopping does not undo actions it has already completed in an external system.

• Read about Tasks and their evidence